A system that integrates multi-source data to intelligently evaluate and predict the risk of chronic diseases
By integrating the intelligent evaluation and prediction system of multi-source data, using KNN classifier and proportional hazards regression model, the problem of limited sample size is solved, and high accuracy assessment and prediction of chronic disease risk is achieved, through Spearman rank correlation detection and gradient The decreasing method optimizes parameters to improve prediction accuracy.
Patent Information
- Application Number
- CN202111070025.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2021-09-13
- Publication Date
- 2024-12-31
- Estimated Expiration
- 2041-09-13
AI Technical Summary
Due to the limited sample size, the prior art cannot accurately evaluate and predict the risk of chronic diseases, resulting in the scarcity of target users' data.
A converged multi-source data intelligent evaluation and prediction system is proposed. Through data acquisition, factor search, risk mapping, data training verification and data evaluation prediction modules, KNN classifier and proportional hazard regression model are used to train the KNN classifier until the classification error reaches the threshold. , establish a mapping relationship between multi-source data and chronic disease pathogenic risk factors, and conduct accurate evaluation.
High accuracy classification and chronic disease risk assessment of multi-source data are achieved, and parameters are optimized through Spearman's level correlation detection and gradient descent method to reduce errors and improve prediction accuracy.
Smart Images
Figure CN113921136B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of artificial intelligence, and in particular to a system for intelligently evaluating and predicting chronic disease risks by integrating multi-source data. Background Art
[0002] The full name of chronic disease is chronic non-communicable disease. It does not refer to a specific disease, but a general term for a class of diseases with hidden onset, long course and prolonged illness, lack of definite infectious biological etiology evidence, complex etiology, and some diseases that have not yet been fully confirmed. The occurrence of chronic diseases is often not caused by a single risk factor, but is often the result of the combined action of multiple risk factors. A chronic disease is often the result of the combined action of multiple risk factors. At the same time, a risk factor can also lead to an increased risk of multiple chronic diseases. For example, the effects of long-term drinking: 1) Liver damage: fat accumulation in the liver causes fatty liver; 2) Gastric ulcer: can cause gastric bleeding and endanger life; 3) Nervous system damage: such as peripheral neuropathy; 4) Cerebral cortex atrophy: chronic alcoholics have atrophy of the cerebral cortex, and some patients also have signs of intellectual decline; 5) Alcoholic fetus: alcohol is metabolized and excreted slowly in the fetus, causing various injuries to the developing fetus, including fetal malformations, fetal death, growth retardation and behavioral defects.
[0003] Therefore, if we can effectively utilize the sample size of the multi-source data above and integrate it into certain indicator evaluation, we can solve the problem of obtaining and evaluating the risk of chronic diseases within a limited sample size and a credible range. Based on this, a system that integrates multi-source data to intelligently evaluate and predict the risk of chronic diseases has found a new breakthrough in constructing a proportional risk regression model of multi-source information and data and a global calculation method. Summary of the invention
[0004] In view of this, the present invention proposes a system for intelligently evaluating and predicting chronic disease risks by integrating multi-source data, aiming to solve the problem that the existing technology is usually unable to accurately evaluate and predict due to limited sample size and scarce target user data. The present invention achieves high-accuracy chronic disease prediction and evaluation of multi-source data by training the KNN classifier until its classification error reaches a threshold.
[0005] The technical solution of the present invention is achieved in this way:
[0006] The present invention proposes a system for intelligently evaluating and predicting the risk of chronic diseases by integrating multi-source data, which includes the following modules:
[0007] Data collection module, to obtain multi-source data of target users;
[0008] The factor search module searches the database of direct and indirect risk factors for chronic diseases and the chronic disease knowledge base, searches the database and the knowledge base, and uses the searched data as the direct and indirect risk factor data for chronic diseases;
[0009] A risk mapping module, based on the direct and indirect risk factor data of chronic diseases, establishes a mapping relationship model corresponding to multi-source data and direct and indirect risk factors of chronic diseases, and collects mapping data according to the mapping relationship model as a mapping data set;
[0010] A data training and verification module uses the mapping data set as an input of an improved KNN classifier to train the KNN classifier until its classification error reaches a threshold;
[0011] The data evaluation and prediction module establishes a proportional risk regression model based on the classified data set to evaluate the relative risk indicators of chronic diseases of target users.
[0012] Preferably, in the data acquisition module, the multi-source data include behavioral data: including basic information, personal medical history, family medical history, smoking and drinking habits, dietary habits, sleeping habits and other data; biochemical indicator data: including blood routine, urine routine, stool routine, liver function, routine electrocardiogram, internal medicine examination, surgical examination and other data; other data, etc.
[0013] Preferably, in the factor search module, after the search related data is used as the direct and indirect risk factor data of the chronic disease, the direct and indirect risk factor data of the chronic disease are normalized.
[0014] On the basis of the above technical solutions, it is further preferred that the direct and indirect risk factor data for chronic diseases include, but are not limited to, genetic factors, gender and age factors, occupational factors, dietary habit factors and overweight factors.
[0015] Specifically, in the data training and verification module, the training of the KNN classifier includes the following units:
[0016] A distance calculation unit calculates the distance between each point in the known category mapping data set and the input test point;
[0017] Sorting units,sorted in increasing order of distance;
[0018] Select the unit, select the k points closest to the input test point;
[0019] A frequency determination unit determines the frequency of occurrence of the categories of the first k points;
[0020] The prediction classification unit returns the category with the highest occurrence frequency of the first k points as the predicted classification of the input point to be tested.
[0021] Preferably, in the distance calculation unit, before calculating the distance between each point in the known category mapping data set and the input test point, the Spearman rank correlation is used to perform correlation detection on every two mapping data in the same disease category in the mapping data set, and multiple groups of mapping data with correlation higher than a preset correlation threshold are retained as the first sample set for KNN classification; multiple groups of mapping data with correlation lower than the preset correlation threshold are retained as the second sample set for further KNN classification.
[0022] Preferably, in the sorting unit, before sorting in ascending order of distance, the calculated distances between points in each group of mapping data are weighted to divide each group of mapping data into groups with different weights.
[0023] Preferably, in the selection unit, before selecting k points closest to the input test point, a cross-validation method is used to obtain the k value corresponding to the KNN classification of the first sample set, and the k value is used as the k value for further KNN classification of the subsequent second sample set.
[0024] Preferably, the data training verification module also includes an error estimation unit, which is used to set an error threshold and perform error estimation on the prediction result returned by the KNN classifier. If the error threshold is not reached, the KNN classifier is repeatedly trained.
[0025] Preferably, in the data evaluation and prediction module, after a proportional risk regression model is established based on the classified data set, the gradient descent method is used to estimate the parameters.
[0026] The system of the present invention for intelligently evaluating and predicting chronic disease risks by integrating multi-source data has the following beneficial effects compared with the prior art:
[0027] (1) The present invention realizes the classification of multi-source data by training the KNN classifier. Compared with the traditional K nearest neighbor algorithm, this method uses Spearman rank correlation to perform correlation detection, thereby excluding a lot of irrelevant data, and then weighting the calculated distance, giving priority to dividing the weights of different distances, and then further completing the classification process. Finally, according to the result of the correlation detection, the optimal classification scheme can be adopted according to the k value corresponding to the high correlation obtained by the correlation detection, thereby ensuring the smallest possible error in the classification process;
[0028] (2) After establishing the proportional risk regression model based on the classified data set, the present invention also uses the gradient descent method to estimate the parameters. Through multiple iterations of the algorithm, the minimum and / or maximum values are calculated step by step and the approximate coordinate points are connected, so as to analyze and calculate a set of linear regression equations that can be used to estimate the prediction. This method gives an estimated value, which is convenient for comparison and analysis with the estimated value after the experimental value is obtained. BRIEF DESCRIPTION OF THE DRAWINGS
[0029] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.
[0030] Figure 1 This is a system structure diagram of the present invention for integrating multi-source data to intelligently evaluate and predict chronic disease risks;
[0031] Figure 2 Schematic diagram of the unit structure of the data training and verification module 40 of the present invention. DETAILED DESCRIPTION
[0032] The following will be combined with the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0033] like Figure 1 As shown, the present invention is a system for intelligently evaluating and predicting chronic disease risks by integrating multi-source data, including the following modules: a data acquisition module 10, a factor search module 20, a risk mapping module 30, a data training and verification module 40 and a data evaluation and prediction module 50. The functions implemented by the specific modules can be referred to the invention content section of the specification and will not be elaborated here.
[0034] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.
[0035] The present invention provides a specific embodiment of a system for intelligently assessing the risk of hypertension by fusing multi-source data. The following is the specific content of the embodiment.
[0036] The data collection module 10 obtains multi-source data of the target user.
[0037] Specifically, the multi-source data of target users include: 1) behavioral data: including basic information, personal medical history, family medical history, smoking, drinking, dietary habits, exercise, sleep, living environment factors and other data; 2) routine / biochemical indicator data: including general examination, internal medicine, surgery, ophthalmology, dentistry, gynecology, otolaryngology, blood routine, urine routine, stool routine, liver function, kidney routine, blood lipids, hepatitis B, tumor marker examination, routine electrocardiogram, X-ray examination and other data; 3) other data, etc.
[0038] The factor search module 20 searches the database of direct and indirect risk factors for chronic diseases and the chronic disease knowledge base, traverses the database and the knowledge base, and uses the searched related data as the direct and indirect risk factor data for chronic diseases.
[0039] Specifically, we can start with the common direct and indirect risk factors for hypertension. For example, the factors include the following aspects: 1) Related to genetic factors. If it is familial, such as both parents are hypertensive patients, the next generation is more likely to suffer from hypertension, and the incidence rate of their children can be as high as 46%; 2) Related to gender and age. There are more male hypertensive patients than female patients, and with increasing age, the number of hypertensive patients also increases. Therefore, hypertensive patients are more common in middle-aged and elderly people; 3) Related to occupation. For people who engage in mental work for a long time or are under long-term high mental stress, the risk of developing hypertension will be higher than that of physical laborers; 4) Related to eating habits. Long-term high-salt diet, drinking, smoking, and drinking coffee will significantly increase the risk of hypertension; 5) Related to overweight and obesity. The risk of hypertension in obese patients is 2-6 times that of people with normal weight.
[0040] Specific relevant experimental data include, but are not limited to, salt intake is positively correlated with blood pressure level and hypertension prevalence, while potassium salt intake is negatively correlated with blood pressure level. Studies have shown that an average daily increase of 2 grams of sodium salt will increase systolic and diastolic blood pressure by 2.0 mmHg and 1.2 mmHg respectively. Excessive drinking is also a risk factor for the onset of hypertension. Although blood pressure will drop in a short period of time after drinking a small amount of alcohol, long-term drinking of small amounts of alcohol can cause a slight increase in blood pressure, while excessive drinking will significantly increase blood pressure. For example, if an average of more than 3 standard cups of alcohol are consumed per day (1 standard cup is equivalent to 12 grams of alcohol, about 360 grams of beer, or 100 grams of wine, or 30 grams of liquor), systolic and diastolic blood pressure will increase by an average of 3.5 mmHg and 2.1 mmHg respectively, and the increase in blood pressure increases with the amount of alcohol consumed. According to studies, people with BMI ≥ 24 have a 3-4 times higher risk of developing hypertension than people with normal weight, and those with a waist circumference of ≥ 90 cm for men or ≥ 85 cm for women have a risk of developing hypertension that is more than 4 times that of people with normal waist circumference.
[0041] In the factor search module 20, after using the successfully searched data as the direct and indirect risk factor data for chronic diseases, the following steps are also included to normalize the direct and indirect risk factor data for chronic diseases.
[0042] Specifically, the normalization process includes the following steps: the direct and indirect risk factors for hypertension are sorted from small to large according to the investigated experimental data, and the influence degree of the experimental data is artificially stipulated so that the value corresponding to the influence degree of the experimental data falls between [0,1]. The conversion function used in the normalization process is x*=(x-min) / (max-min), wherein x* represents the converted data, x represents the experimental data, min represents the value corresponding to the smallest influence degree, and max represents the value corresponding to the largest influence degree.
[0043] The risk mapping module 30 establishes a mapping relationship model corresponding to multi-source data and direct and indirect risk factors of chronic diseases based on the direct and indirect risk factor data of chronic diseases, and collects mapping data according to the mapping relationship model as a mapping data set.
[0044] Specifically, establishing a mapping relationship model includes the following steps. For example, a blood pressure check is performed on the target user, and the systolic pressure and diastolic pressure obtained are part of the multi-source data. At the same time, combined with the target user's oral drinking habits, the mapping relationship between drinking habits and hypertension can be reflected in the model based on the research results described above.
[0045] The data training and verification module 40 uses the mapping data set as the input of the KNN classifier and trains the KNN classifier until its classification error reaches a threshold.
[0046] Specifically, in the data training verification module 40, the training KNN classifier includes the following units:
[0047] A distance calculation unit calculates the distance between each point in the known category mapping data set and the input test point;
[0048] Sorting units,sorted in increasing order of distance;
[0049] Select the unit, select the k points closest to the input test point;
[0050] A frequency determination unit determines the frequency of occurrence of the categories of the first k points;
[0051] The prediction classification unit returns the category with the highest occurrence frequency of the first k points as the predicted classification of the input point to be tested.
[0052] Preferably, in the distance calculation unit, before calculating the distance between each point in the known category mapping data set and the input test point, the Spearman rank correlation is used to perform correlation detection on every two mapping data in the same disease category in the mapping data set, and multiple groups of mapping data with correlation higher than a preset correlation threshold are retained as the first sample set for KNN classification; multiple groups of mapping data with correlation lower than the preset correlation threshold are retained as the second sample set for further KNN classification.
[0053] Preferably, in the sorting unit, before sorting in ascending order of distance, the calculated distances between points in each group of mapping data are weighted to divide each group of mapping data into groups with different weights.
[0054] Preferably, in the selection unit, before selecting k points closest to the input test point, a cross-validation method is used to obtain the k value corresponding to the KNN classification of the first sample set, and the k value is used as the k value for further KNN classification of the subsequent second sample set.
[0055] On the basis of the above technical scheme, preferably, according to the research data on hypertension caused by excessive sodium salt consumption mentioned above, the risk of hypertension caused by excessive sodium salt consumption is 1.16-1.28 times that of normal people, and the research data on hypertension caused by excessive drinking mentioned above shows that the risk of hypertension caused by excessive drinking is 1.5-2 times that of normal people. According to the research data, excessive drinking is more closely related to the causative effect of hypertension. In order to ensure the accuracy and authenticity of the data, it is also necessary to perform testing through Spearman rank correlation.
[0056] On the basis of the above technical solution, preferably, the distance between each group of mapping data points after calculation is weighted, and each group of mapping data is divided into different weights. The weighting method adopts Gaussian function weighting. As the distance increases, the weight will decrease, and the weight will not become 0, avoiding data export errors due to too low weight.
[0057] It should be understood that, since the samples are classified in advance by correlation, the samples with high correlation can be used as high-risk factors for a single chronic disease to a certain extent, while for the mapping data with low correlation, the selection of k value needs to be guaranteed. That is to say, first, multiple groups of mapping data with correlation higher than the preset threshold should be extracted for preliminary classification to determine the optimal k value. Then, the value of k is used as the k value for all subsequent groups of mapping data with correlation lower than the preset correlation threshold, and the test points are classified again. This is to avoid the error of k value of mapping data with low correlation being too large compared with that of mapping data with high correlation when performing KNN classification. The two classification results are compared and the classification result with the highest classification accuracy is taken as the final disease prediction result.
[0058] On the basis of the above technical solution, preferably, in the data training verification module 40, during the process of training the KNN classifier, an error threshold is set, and the prediction result returned by the KNN classifier is subjected to error estimation, and the KNN classifier is repeatedly trained if the error threshold is not reached.
[0059] The data evaluation and prediction module 50 establishes a proportional risk regression model based on the classified data set to evaluate the relative risk index of chronic diseases of the target users.
[0060] After using the KNN classification algorithm to classify the mapping relationship between multi-source data and the direct and indirect risk factors of chronic diseases, each data set can be plotted according to the coordinate points and a proportional risk regression model can be established, and distinguished by lines of different colors. This will make the evaluation of the relative risk index of hypertension for the target user more intuitive and clear.
[0061] On the basis of the above technical solutions, preferably, in the data evaluation and prediction module 50, after the proportional risk regression model is established according to the classified data set, the following steps are also included, in which the parameters are estimated by the gradient descent method, the main purpose of which is to find the minimum value of the objective function through iteration, or converge to the minimum value, for example, according to the parameter data of various factors in the proportional risk regression model, each parameter is further estimated for the risk of hypertension. In general, the more factors there are, the higher the probability of disease, and under the influence of the same factors, such as drinking habits, the increase in blood pressure increases with the increase in drinking volume, and the specific experimental data requires the use of the gradient descent method to make an accurate estimate of the parameters.
[0062] Contents not described in detail in this specification, such as mathematical formulas used in various algorithms, belong to the prior art known to professional and technical personnel in this field.
[0063] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.
Claims
1. A system for intelligently evaluating and predicting the risk of chronic diseases by integrating multi-source data, characterized in that: Used to assess the risk of hypertension, including the following modules: A data collection module is used to obtain multi-source data of the target user. In the data collection module, the multi-source data includes behavioral data and biochemical index data. The behavioral data includes basic information, personal medical history, family medical history, smoking and drinking habits, dietary habits and sleeping habits; the biochemical index data includes blood routine, urine routine, stool routine, liver function data, routine electrocardiogram, internal medicine examination and surgical examination data; A factor search module searches the database of direct and indirect risk factors for chronic diseases and the chronic disease knowledge base, traverses and searches the database and the knowledge base, uses the searched data as the direct and indirect risk factor data for chronic diseases, and normalizes the direct and indirect risk factor data for chronic diseases, wherein the direct and indirect risk factor data for chronic diseases include genetic factors, gender and age factors, occupational factors, dietary habit factors, and overweight factors; A risk mapping module, based on the direct and indirect risk factor data of chronic diseases, establishes a mapping relationship model corresponding to multi-source data and direct and indirect risk factors of chronic diseases, and collects mapping data according to the mapping relationship model as a mapping data set; A data training and verification module uses the mapping data set as an input of an improved KNN classifier to train the KNN classifier until its classification error reaches a threshold; In the data training and verification module, the training of the KNN classifier includes: The distance calculation unit uses Spearman rank correlation to perform correlation detection on every two mapping data in the same disease category in the mapping data set, retains multiple groups of mapping data with correlation higher than a preset correlation threshold as the first sample set, and performs KNN classification; multiple groups of mapping data with correlation lower than the preset correlation threshold are used as the second sample set, and further performs KNN classification; calculates the distance between each point in the known category mapping data set and the input test point; A sorting unit weights the distances between each group of mapping data points after calculation, divides each group of mapping data into groups with different weights, and sorts them in ascending order of distance; Select the unit, use the cross-validation method to obtain the k value corresponding to the KNN classification of the first sample set, and use it as the k value for the subsequent KNN classification of the second sample set; select the k points closest to the input test point; A frequency determination unit determines the frequency of occurrence of the categories of the first k points; The prediction classification unit returns the category with the highest occurrence frequency of the first k points as the predicted classification of the input point to be tested; The data training verification module also includes an error estimation unit, which is used to set an error threshold and perform error estimation on the prediction result returned by the KNN classifier. If the error threshold is not reached, the KNN classifier is repeatedly trained. The data evaluation and prediction module establishes a proportional risk regression model based on the classified data set to evaluate the relative risk indicators of chronic diseases of target users.
2. The system for intelligently evaluating and predicting chronic disease risks by integrating multi-source data as claimed in claim 1, characterized in that: In the data evaluation and prediction module, after the proportional hazards regression model is established based on the classified data set, the gradient descent method is used to estimate the parameters.